How to Run Ollama on a Synology NAS — Setup Guide 2026

Ollama on a Synology NAS runs an open-source large language model entirely on your own hardware, accessible to every device on your local network, with no cloud service and no per-token cost. The setup uses Synology's Container Manager (Docker) to run Ollama as a container, which exposes a local API and optionally Open WebUI for a browser-based chat interface. This guide covers every step from hardware check to first conversation, including the specific RAM and CPU requirements that determine which Synology models are actually viable. For a hardware requirements overview, see AI NAS Hardware Requirements.

Compatible Synology Models for Ollama

Not all Synology NAS units can run Ollama. The requirements are: x86 processor, DSM 7.x, Container Manager installed, and sufficient RAM. ARM-based Synology units (all J-series, base DS423) cannot run Ollama because llama.cpp requires x86 instruction support.

Recommended Synology models for Ollama (AU retail):

• DS925+ (from $980): AMD Ryzen R1600, expandable to 32 GB. Best current 4-bay Synology for AI. Upgrade to 16 GB RAM before setting up Ollama.
• DS1825+ (from $1,699): AMD Ryzen V1780B, 4-core, expandable to 32 GB. Faster than DS925+ for AI inference due to higher core count.
• DS1525+ (from $1,233): AMD Ryzen based. Expandable RAM. Good option if already owned.
• DS425+ (from~$1199): Intel Celeron N95, expandable to 6 GB only. Limited to 3B models. Not recommended for 7B inference due to RAM ceiling.

Models that will not work: DS223J, DS423, any J-series or value-tier Synology. If unsure about your model's CPU, check Synology's spec page for the processor type. "Realtek" or "Cortex-A" in the CPU field = ARM = Ollama not viable.

Step 1: Install Container Manager

Container Manager is Synology's Docker runtime. It replaced Docker for DSM 7.2 and later. If you are running DSM 7.1 or earlier, update DSM before proceeding.

1. Open Package Center in DSM.
2. Search for Container Manager.
3. Click Install. The package is free.
4. Once installed, open Container Manager from the main menu.

If Container Manager is greyed out or unavailable, your NAS model does not support it (likely ARM). This is the definitive compatibility check: if Container Manager installs, your NAS can run Ollama.

Step 2: Create Folders for Ollama Data

Ollama stores model weights on disk. Create a dedicated folder for this before deploying the container.

1. Open File Station in DSM.
2. Navigate to your main volume (typically volume1).
3. Create a folder named docker (if it does not already exist).
4. Inside docker, create a subfolder named ollama.
5. Inside docker, create a second subfolder named open-webui (for later).

The path will be /volume1/docker/ollama. This is where model weights are stored. If you have an NVMe M.2 SSD installed as a cache volume, store model weights there instead for faster model loading.

Step 3: Deploy the Ollama Container

You can deploy Ollama via Container Manager's Project feature using Docker Compose, which is the cleanest approach for adding Open WebUI later.

1. Open Container Manager and select Project from the left sidebar.
2. Click Create.
3. Name the project ollama.
4. Set the path to /volume1/docker/ollama.
5. Select Create docker-compose.yml and paste the following:

version: '3.8'
services:
  ollama:
    image: ollama/ollama:latest
    container_name: ollama
    restart: unless-stopped
    ports:
      - "11434:11434"
    volumes:
      - /volume1/docker/ollama:/root/.ollama
    environment:
      - OLLAMA_HOST=0.0.0.0

  open-webui:
    image: ghcr.io/open-webui/open-webui:main
    container_name: open-webui
    restart: unless-stopped
    ports:
      - "3000:8080"
    volumes:
      - /volume1/docker/open-webui:/app/backend/data
    environment:
      - OLLAMA_BASE_URL=http://ollama:11434
    depends_on:
      - ollama

Click Next and then Done. Container Manager will pull the Ollama and Open WebUI images, which may take 5-10 minutes depending on your internet connection. Both containers start automatically once pulled.

If you prefer to run Ollama alone without Open WebUI, deploy only the ollama service block. Open WebUI can be added later as a separate project.

Step 4: Download Your First Model

With Ollama running, pull a model via the Container Manager terminal or via SSH.

Option A: Container Manager terminal.

1. In Container Manager, click Container in the left sidebar.
2. Click the ollama container.
3. Click the Terminal tab.
4. Click Create to open a shell session.
5. Run: ollama pull llama3.2:3b

Option B: SSH.

1. Enable SSH in DSM Control Panel.
2. SSH into the NAS: ssh admin@your-nas-ip
3. Run: docker exec -it ollama ollama pull llama3.2:3b

Recommended first models for Synology hardware:

• llama3.2:3b. 2 GB download, runs on 6 GB+ RAM, fast on DS925+ hardware
• phi3:mini. 2.2 GB download, efficient on limited hardware, strong for its size
• llama3.2:7b. 4.7 GB download, requires 16 GB RAM configuration, best quality at this tier
• mistral:7b. 4.1 GB download, strong for coding and structured tasks, 16 GB RAM

Model names are case-sensitive. Run ollama list to see what is currently installed on your NAS.

Step 5: Access Open WebUI and Start Chatting

Open a browser on any device on your local network and navigate to http://your-nas-ip:3000. The Open WebUI interface loads, prompting you to create an admin account on first access. This account is local to your NAS and is not connected to any external service.

Once logged in, select a model from the dropdown at the top of the chat interface. The models you pulled in Step 4 appear here. Start a conversation: your prompt is sent to the Ollama API on the NAS, processed by the model using the NAS CPU, and the response is streamed back through the WebUI.

Open WebUI supports conversation history, document uploads for context (PDF, text), system prompt customisation, and multiple model switching within a session. For most home and SMB use cases, this provides a complete private AI chat environment.

Performance Expectations on Synology Hardware

Synology NAS hardware is optimised for storage, not AI compute. Setting accurate expectations prevents frustration.

DS925+ (AMD R1600, 2-core, 16 GB RAM):

• Llama 3.2 3B: approximately 5-8 tokens per second
• Llama 3.2 7B: approximately 2-4 tokens per second
• A 200-word response takes approximately 40-70 seconds at 7B

DS1825+ (AMD V1780B, 4-core, 16 GB RAM):

• Llama 3.2 3B: approximately 8-12 tokens per second
• Llama 3.2 7B: approximately 4-6 tokens per second

These speeds are adequate for document summarisation, private Q&A, and background processing tasks. They are noticeably slower than cloud AI for interactive conversation. If real-time conversational speed is required and a GPU PCIe slot is available, see the notes on GPU acceleration for QNAP PCIe-capable models in AI NAS Hardware Requirements. For QNAP-based Ollama setup with AMD Ryzen hardware, see Ollama on QNAP NAS.

Running Ollama has a measurable power impact: the NAS draws 10-20W more during inference than at idle. Use the NAS Power Cost Calculator to model annual electricity costs. For general NAS selection advice, see Best NAS Australia.

Troubleshooting Common Issues

Container fails to start. Check that ports 11434 (Ollama) and 3000 (Open WebUI) are not already in use by another Synology service. In DSM, check Resource Monitor for port conflicts. If Synology's built-in firewall is enabled, add rules to allow these ports on the local network.

Model pulls fail with network error. Ollama downloads models from Meta, Mistral AI, and other open-source registries. If your network has content filtering, whitelist the Ollama model registry domains. Models can also be manually downloaded on another machine and copied to the /volume1/docker/ollama folder in the correct directory structure.

Inference is very slow or causes NAS to freeze. The most common cause is insufficient RAM, forcing the NAS to use swap (disk-backed virtual memory). Check RAM usage in Resource Monitor during inference. If RAM is at or near 100%, reduce the model size (switch from 7B to 3B) or add a physical RAM upgrade. Swap-backed LLM inference is orders of magnitude slower than RAM-backed inference.

Open WebUI cannot connect to Ollama. Verify both containers are running in Container Manager. The OLLAMA_BASE_URL environment variable in the Open WebUI container must point to the Ollama container. If using the Docker Compose approach above, the service name ollama resolves correctly within the Docker network. If deployed as separate projects, use the NAS IP address instead: http://your-nas-ip:11434.

Australian Buyers: What You Need to Know

Hardware sourcing in AU. The Synology DS925+ (from $980) is available at Mwave, PLE Computers, Scorptec, and Computer Alliance. RAM upgrades (DDR4 SO-DIMM non-ECC) are available from the same retailers and MSY. A 16 GB SO-DIMM stick suitable for the DS925+ costs approximately $55-80. Check Synology's official compatibility list for confirmed RAM modules.

Privacy benefit. Running Ollama on a Synology NAS means your prompts and documents are never transmitted to any external server. For Australians processing personal or business data, this avoids any Privacy Act 1988 cross-border disclosure concerns that arise from cloud AI services. Your data stays on hardware you own, in your location.

Ollama is not officially supported by Synology. Container Manager supports Docker containers broadly, and Ollama runs as a Docker container, so it functions correctly. However, Synology does not test or certify Ollama specifically. Support for Docker-based applications comes from the community, not Synology's support team. If you raise a DSM issue unrelated to Ollama and mention it in the support ticket, Synology may note the unsupported configuration.

NBN and model downloads. Pulling a 7B model for the first time downloads approximately 4-5 GB. On NBN 100 (typical 100 Mbps download), this takes approximately 5-10 minutes. Model weights are stored locally after download and do not require re-downloading. Australian Consumer Law protections apply to Synology hardware purchased from AU retailers.

Related reading: our Synology brand guide, our NAS vs cloud storage comparison, and our NAS explainer.

Use our free AI Hardware Requirements Calculator to size the hardware you need to run AI locally.

Does Ollama work on all Synology NAS models?

No. Ollama requires an x86 processor. Synology units with ARM processors (J-series, most value-tier DS models, DS423) cannot run Ollama. x86 Synology units with Container Manager support can install and run Ollama. The DS425+, DS925+, DS1525+, and DS1825+ are the recommended current AU models. Always verify the processor type on Synology's spec page before purchasing for AI use.

Can I access Ollama on my Synology NAS from outside my home network?

Yes, via VPN. Synology VPN Server (OpenVPN, WireGuard) allows access to the entire local network remotely, including the Ollama API and Open WebUI. Do not expose port 11434 (Ollama) directly to the internet without authentication middleware, as Ollama's API has no built-in authentication by default. Open WebUI includes user authentication and is safer to expose, but VPN is the recommended approach for security. See NAS Remote Access via VPN for a full setup guide.

How much disk space do LLM models take on a Synology NAS?

Model sizes in 4-bit quantization (the default Ollama download): 3B models take approximately 2 GB, 7B models approximately 4.5 GB, 13B models approximately 8 GB. Running 3-4 different 7B models requires 15-20 GB of dedicated storage. An NVMe M.2 SSD (if your Synology model has an M.2 slot) is the best location for model storage, providing faster model loading times than HDD-backed volumes.

Will running Ollama affect my NAS backup and storage performance?

During active inference, CPU usage increases substantially. Simultaneous backup tasks (scheduled backups, HyperBackup) competing for CPU will both slow down. Schedule CPU-intensive tasks (Ollama inference, backups, thumbnail generation) to avoid overlap. Synology's Resource Monitor shows real-time CPU and RAM usage, which helps identify contention. Overnight batch AI processing and overnight backup schedules should be staggered by 1-2 hours.

What is the difference between Ollama and Synology AI Console?

Synology AI Console is Synology's own AI integration, built into DSM and designed for photo, document, and productivity features. It uses models chosen and maintained by Synology, runs seamlessly with other DSM apps, and requires no user configuration. Ollama is an open-source tool you install yourself that runs any compatible open-source model. They serve different purposes: AI Console is the polished vendor experience; Ollama is the flexible, any-model-you-want approach. They can run on the same NAS simultaneously without conflict.